On Jul 10, 2015 6:10 AM, "Michael Rasmussen" <[email protected]> wrote: >
> > Given your chain, each node consumes data from the following one. > Node I generates the raw data. The system is designed on a pull > model. If C fails B and A think there's no work to do - and eventually > generate alarms because they expect there always should be something > to do. D will write to local storage. When C is brought back it will > process the queued data on D. > > Aside from getting disk arrays, databases and message queuing services started > there are no actual hard dependencies. The desire to control and coordinate > startups is threefold. > > - ensure health of component before going forward > - not populate logs with messages about lack of services > - not generate a backlog of data to be processed when the full system is up. > > If instead of a pull model there was a push model then the pushing node could start the downstream service when it had something to push. Nothing would be started until it was needed the first time. Also if it can't start or find anything to push to it can send a message upstream pausing the rest of the nodesl much like the kill switch in the Toyota production line. No matter how finely it is tuned one of the nodes will be a bottleneck and will have to pause the upstream line occasionally even under normal conditions. Bill > -- > Michael Rasmussen, Portland Oregon > Be Appropriate && Follow Your Curiosity > Complaining is silly. Either act or forget. > ~ Sagmeister > _______________________________________________ > PLUG mailing list > [email protected] > http://lists.pdxlinux.org/mailman/listinfo/plug _______________________________________________ PLUG mailing list [email protected] http://lists.pdxlinux.org/mailman/listinfo/plug
